sMRI Classification of Alzheimer's Disease Using Genetic Algorithm and Multi-Instance Learning (GA+MIL)

Text, Image, Audio and Video are the data formats used for storing information in today's digital world. In particular, Image processing plays a crucial part in the medical profession for diagnosis of diseases and severity identification, and the modality of input can be either 2D or 3D images. A type of disease named Alzheimer's disease (AD) occurs due to damage to nerve cells (neurons) in the human's brain. Alzheimer's disease is regarded as one of the main causes of death worldwide. Alzheimer's disease diagnosis is a complex task, and no preventive measures have been ascertained yet. This study aims to create a system that uses 3D structural magnetic resonance imaging (sMRI) as an input for the diagnosis of Alzheimer's disease. Unlike existing architecture like basic Convolutional Neural Networks (CNN), the methodology Genetic Algorithm + Multi-Instance Learning (
$\text{GA} +\text{MIL}$
) is an advanced model that works on two levels, Local Feature Extraction and Global Feature Extraction is used for this research. In general, the partition of Region of Interests (ROI) from sMRI images is basically divided into three categories namely, 1) Voxel (volumetric pixel)-level, 2) Region-level and 3) Patch-level methods. Correlation and overfitting problem is a major drawback in Voxel-level method whereas Region-level method has the disadvantage of not including all the disease related features. To overcome this problem, the Patch-level method of partition of ROIs is adopted in this study. Hence the initial sMRI is preprocessed, then divided into patches and passed into the
$\text{GA} +\text{MIL}$
model. The experimental results show that the usage of Genetic Algorithm to identify discriminative pathological locations results in efficient global feature extraction.